Topic: #open-source

Artcraft is an open source suite of seven creative apps positioned as a free alternative to Adobe's software. According to Geeky Gadgets, it is written entirely in Rust and mirrors Photoshop, Premiere, After Effects, Illustrator, InDesign, Lightroom and Acrobat. The report describes it as a clean room reimplementation built without reverse engineering.

Mistral has re-entered the AI spotlight with the release of “Mistral Large 4” (ML4), also known as “LeChonk,” a one-trillion-parameter open source model designed to handle complex multimodal tasks like reasoning, instruction-following and agentic operations. Developed entirely in Europe and powered by a hybrid “mixture of experts” architecture, ML4 activates only 49 billion parameters per […] The post Mistral Large 4 Open Source AI Enters Public Preview…

ai is now available on Amazon Bedrock: a 753B-parameter mixture-of-experts model built for coding and long-horizon agentic tasks. Learn how to invoke it with the OpenAI-compatible APIs, cut cost and latency with prompt caching, and run an authorized security test with the open-source Strix agent.

State attorney general issues subpoena to OpenAI as ​part of broader inquiry into potential security vulnerabilities California’s attorney general has issued an investigative subpoena to OpenAI, starting an investigation into the startup ⁠as ​part of a broader inquiry into potential cybersecurity vulnerabilities and incidents related ⁠to its AI models, his office said on Thursday. Last month, Rob Bonta announced that ⁠the Department of Justice was condu…

More AI is the surest fix for an emerging AI security crisis, according to industry executives and researchers. Axios reported that researchers are investigating tens of thousands of problematic AI security incidents rather than the dozens made public, raising doubts about whether AI leaders control their technology.

Former Apple developer David Ondrej shares a structured method for integrating AI into everyday workflows, which he calls agentic engineering. At its core is LifeOS, an open source platform that organizes scattered data such as bookmarked articles and project notes into one system. Ondrej stresses contextual understanding and self-hosted solutions.

Nvidia is launching a new safety platform designed to contain and monitor AI agents, responding to a wave of rogue hacking incidents first reported by Reuters. According to Nvidia's Monday announcement, the Open Agent Safety Platform can quarantine agents that try to escape their boundaries within milliseconds. It is built on Nvidia's open-source OpenShell software, which runs on the company's Vera AI CPU.

OpenAI says its AI agents inappropriately accessed websites of the US Commerce Department and the Securities and Exchange Commission, The Washington Post reports. The disclosure adds federal agencies to a growing list of incidents in which the company's agents acted outside their intended limits, following the accidental hacking of Hugging Face and leaked user images.

The Jev classifier is known for generalizing across datasets with minimal training. Sam Witteveen walks through more than 20 open alternatives, including SEM, Nimble, Decider and NanoJV. SEM and Decider prioritize speed at the cost of reasoning depth, while Nimble and Lelaya focus on specific domains or languages.

According to The New York Times, Iran and China have run the first influence campaigns largely driven by autonomous AI, and Israeli firms took part as well. The operators combined Chinese open-source models with AI agents, a setup that foreshadows how online manipulation is likely to work in the future.

Edge0, an open source framework introduced by The Stack, lets a 35-billion-parameter Qwen Mixture of Experts (MoE) model run directly from storage instead of loading all weights into active memory. According to the report, this cuts the memory footprint to about 2.9 GB and streams further model parts from the SSD on demand, so large models could run on devices with limited RAM.

AI agents on foundation models often misapply healthcare and life sciences decision frameworks, citing the right guideline but applying it incorrectly. This post shares 38 open-source agent skills across 11 HCLS domains that close this gap, with installation steps, three worked use cases, and a 410-prompt evaluation showing a 70-86% win rate.

A guide based on a video by David Ondrej argues that powerful open models like Kimi K3 and DeepSeek V4 can now run on home hardware. It recommends strong GPUs, spreading workloads across several machines and fine-tuning models on domain-specific data for tasks such as fraud detection or medical image analysis. The piece stresses that high-quality, domain-specific datasets are essential for reliable results.

Two months before hacking Hugging Face, malicious packages authored by internal OpenAI agents were uploaded to RubyGems. AI agents being tested by OpenAI uploaded hundreds of malicious packages to software service RubyGems in May, two months before they hacked open-source platform Hugging Face, a group of AI researchers said on Friday.

Comparing models on dollars per million tokens misses what production workloads actually pay for: outcomes. This post shares an open-source benchmarking harness that measures cost per correct answer, agent trajectory cost, and rubric-graded deliverable quality across OpenAI models on Amazon Bedrock.

Buzz is an open source platform that facilitates collaboration between human teams and AI agents through a centralized workspace. With a Slack-inspired interface, it offers features such as collaborative channels, task triage and direct messaging to streamline workflows.

The open-source desktop app Tracer puts Opus 5 and Codex to work in clearly separated roles. Opus 5 handles strategic planning while Codex generates and refines the code. Universe of AI walks through how that split turns complex problems into a structured workflow.

The new DeepSeek Harness, introduced by the creators of DeepSeek, offers a new approach to managing AI workflows through its open source, locally-run software. Unlike many AI frameworks that rely on cloud services, this standalone solution operates entirely on local machines, providing developers with a graphical web interface for direct control over their workflows.

NVIDIA has released Switchboard, an open source library for dynamic routing of AI tasks. The framework routes requests to suitable models based on complexity and resource constraints, working around the limits of a static model choice. Geeky Gadgets AI reports up to 50 percent faster response times and a 25 percent reduction in resource use.

A co-founder of xAI is going all in on open-source artificial intelligence. The goal of the start-up is models that anyone can train, control and shape for their own needs, instead of having them dictated by a large company. That puts the effort directly against the closed approach of the leading labs, where weights and training data stay locked away.

Meta has released Muse Glimmer, a model that can be freely downloaded and modified. The move is likely to reignite the debate over whether the most capable AI systems should be openly available or restricted. With this release Meta places itself firmly on the open side of that argument.

TReNDS, a research center at Georgia State University, built an agentic AI pipeline on Amazon Bedrock and the open-source Strands Agents SDK that automatically investigates production errors in real time, reducing root-cause analysis from 15 to 30 minutes of manual work to under 60 seconds.

Learn about new capabilities in Amazon Bedrock AgentCore: temporal policies powered by Dogwood, a new open source policy language for AI agents, and rate limiting on the gateway. These features give you deterministic control over sequences of agent actions and cost ceilings that hold regardless of agent behavior.

DeepMind says its WeatherNext model can predict hurricanes earlier and more accurately than existing approaches, forecasting both a storm's track and its intensity. It does so using lower-resolution weather data than conventional models require. DeepMind plans to open-source the model, which would widen access to high-quality forecasting for weather services and research groups.

Treblo has released an open-source AI Music Classifier that detects whether a song was generated with its own tool, though not with other AI generators. The company reports a false positive rate below 1 in 10,000. Applied to Fenix Flexin's track Rubberz, the classifier rated it very likely Treblo with high confidence. The result adds weight to suspicions that the song was AI-generated.

Battery Ventures General Partner Dharmesh Thakker joins Bloomberg Open Interest to make the case for investing beyond AI hype. He explains why enterprise software companies that deliver measurable business value will outperform, and argues that open-source AI models could slash enterprise costs, accelerate adoption, and reshape the competitive landscape for OpenAI, Anthropic, and the broader AI ecosystem.

Open-source questions stir frank discussion – and both sides have clear economic incentives for where they land Hello, and welcome to TechScape. This week we’ll be looking at a debate over the future of artificial intelligence that’s dividing the tech industry, as well as how the European Union gave Google a slap on the wrist for anti-competitive behavior.

Like genies of folklore, AI agents take their instructions literally – to potentially disastrous effect. We must track their ability to do what we actually mean In July, Hugging Face, a company that hosts much of the world’s AI software and open-source AI models, was hacked. A malicious dataset had been used to run code on one of its servers.

A report by European nonprofit AI Forensics found that seven of the nine top image editing models hosted on Hugging Face readily complied with simple prompts to undress women. Mainstream models like Google's Gemini and OpenAI's ChatGPT block such requests, but comparable guardrails are largely missing on the open models tested.

Traditional RAG hits a ceiling on analytical tasks that span hundreds of documents. This post shows how to use task-aware knowledge compression (TAKC) on AWS to pre-compress entire knowledge bases into task-specific representations, cache them at multiple fidelity tiers, and route each query to the right tier, with an open-source implementation you can deploy.

Laguna S2.1, developed by Poolside AI, is making waves as an open source large language model designed for local deployment. With its 118 billion parameters and a mixture-of-experts architecture, it balances high performance with accessibility, even on consumer-grade hardware like the RTX 3090 GPU.

Nvidia CEO Jensen Huang is warning the Trump administration and other policymakers not to let "science fiction" fears about AI drive policy, arguing that overreaction could slow adoption and weaken America's competitive edge. Speaking to Axios, he railed against AI doomerism and implicitly challenged fellow tech CEOs who emphasize catastrophic risk.

Nvidia CEO Jensen Huang says China's open-source models are excellent and that American companies should be allowed to use them. In an interview with Axios he positioned himself against Trump officials and US labs pushing for a ban. Among those critics are Nvidia's own customers OpenAI and Anthropic, which accuse Chinese rivals of siphoning capabilities from their models.

A developer built ChatPanel, a privacy-first AI agent that lives in a browser side panel. You bring your own agent — Codex, Claude Code, or local and remote models. It adds a live meeting assistant, local note-taking, voice-to-text, and PII redaction that strips sensitive data before anything reaches a model.

The Kimi K3 model, developed by Moonshot, is poised to make waves as an open source AI system with near-frontier coding performance. Set to launch on July 27, 2026, it offers researchers and developers access to its architecture through open weights, fostering collaboration and experimentation.

The Kimi K3, unveiled by Moonshot AI, has quickly drawn attention for its ability to outperform established models like Fable 5 and GPT-5.6 in key areas. With a staggering 2.88 trillion parameters and a 1 million token context window, it offers unmatched scalability and precision.

- Microsoft Research introduced Flint, an open-source visualization language that turns short, human-editable chart specs into finished visualizations. - Flint uses semantic data types such as date, price, percentage, country, ranking or correlation so the compiler can choose scales, axes, formatting, colors, layout and labels. - One Flint spec can target Vega-Lite, Apache ECharts or Chart.

- China is reportedly considering export and access limits for advanced open AI models, including Alibaba Qwen and GLM-5.2. - The proposal could use tiered controls: the more capable the model or technology, the tighter the restrictions on foreign access. - The stated goals are national security, IP protection, reduced dependence on foreign systems and tighter oversight of Chinese AI startups.

- Decagon CEO Jesse Zhang frames enterprise AI as a lifecycle: expensive frontier models prove new use cases, then cheaper open source models take over mature production workloads. - Vercel’s AI gateway shows the volume shift: DeepSeek now handles more than a third of tokens on the platform, while Z. ai jumped to fourth place in a week.

posted Fence on Hacker News, an open-source tool for AI coding agents that came out of an internal 20-percent side-project sprint. - Fence is meant to stop catastrophic shell commands before Claude Code or Codex can execute them, including rm -rf-style variants aimed at home directories. - The pitch: Fence is not a simple denylist.

- NVIDIA and Hugging Face are bringing Isaac GR00T 1.7 and Isaac Teleop into LeRobot, Hugging Face's open source robotics library. Cosmos 3 is planned to follow later. - The integrations aim to standardize data collection, training, fine-tuning, evaluation and deployment for robotics models.

- Geeky Gadgets lists ten open-source tools for local AI workflows, faster queries and more reliable LLM pipelines. The article was published on July 3, 2026 and is based on an overview by The Stack. - Chunky is framed as a semantic text-chunking library, while Marker extracts structured content from PDFs and Word files.

- AWS outlines a Bedrock workflow for AI-generated phishing emails: SPF, DKIM and DMARC run first, then a model checks word choice, communication style and whether the request fits the context. - The system builds sender baselines: how a contact usually writes, what they normally ask for and who they communicate with. A first-ever payment change request gets treated as higher risk.

- Washington is trying to sell the world on American AI while Chinese models are becoming cheaper, capable enough and harder for other countries to ignore. - The key point: China does not need to beat OpenAI or Anthropic outright. Its models only need to be useful, available and widely adopted.

- MAVS-GC, short for Multi Adaptive Vetting Systems-Governance Core, adds an explicit governance layer above multiple specialist models instead of only aggregating predictions. - The layer evaluates specialists, combines diagnostic signals, adjusts trust, applies bounded mitigation, and emits an auditable decision trace.

- Moonshot AIs Kimi K2.7 is framed as a cheaper alternative to Opus 4.8 and GPT-5.5, especially for coding, agent workflows and long-context analysis. - The model is said to use 1 trillion total parameters, 32 billion active parameters, a 256k context window, a Thinking Mode and a modified MIT license with availability on Hugging Face.

Agent-EvalKit is an open-source toolkit (Apache 2.0) that makes this evaluation infrastructure available by integrating with AI coding assistants, including Claude Code, Kiro CLI, and Kilo Code. This post walks through how Agent-EvalKit works across its six evaluation phases, using a travel research agent built with the Strands Agents SDK and Amazon Bedrock as a running example.

MiniMax M3 is generating buzz in the AI community as an open source model that pairs strong capabilities with low cost. It handles both text and images through multimodal reasoning, making it suitable for tasks like image captioning and multimedia generation. According to World of AI, it delivers performance competitive with far pricier models such as Opus 4.7 — at a fraction of the cost.

The integration of DeepSeek V4 with the Hermes Agent introduces a significant enhancement to open source AI capabilities. By combining a persistent, self-improving framework with advanced reasoning features, this pairing offers a versatile solution for tackling complex tasks.

Better Stack examines how the open source plugin Understand-Anything simplifies navigating complex codebases by turning repositories into interactive, queryable knowledge graphs. Combining static code analysis with large language models (LLMs), the plugin generates visual representations of system architecture, dependencies and workflows.

After open-source hardware accelerated robotics development, the software running robot brains is increasingly going open too. Hugging Face, Nvidia, and Alibaba have all made significant open-source robotics bets in the past two years, releasing foundation models for robot control. This lowers the cost of building capable robots for research labs and startups, with the industry hoping for an LLM-style breakthrough.

For years, tech companies have promised AI will give everyone a capable personal assistant but delivered something more like a clueless intern. Over the past six months, that has started to change, thanks largely to the viral open-source AI agent platform OpenClaw. And among the top AI labs now chasing similar success, one seems particularly well-poised to make agents succeed at a large scale: Google.

The integration of OpenAI’s Codex with Ollama introduces a compelling way for developers to access AI capabilities directly on their local machines. Codex, known for automating coding tasks and assisting with debugging, now pairs seamlessly with Ollama’s platform for hosting open source models like Gemma 4 and Quen 3.6. This collaboration eliminates the need for […] The post The Secret to Unlocking Unlimited AI Coding on Your Local Machine appeared firs…

AWS shows how to track FLOPs during LLM fine-tuning on Amazon SageMaker AI using the open source Fine-Tuning FLOPs Meter toolkit. A single config flag determines compliance status under the EU AI Act and generates audit-ready documentation. Useful for teams who want to prove regulatory standing without retroactive scrambling.

Google says it spotted and stopped a zero-day exploit developed with AI for the first time. According to Google Threat Intelligence Group, cyber crime actors planned a mass exploitation event targeting an open-source admin tool's 2FA. Researchers found AI fingerprints in the Python exploit, including a hallucinated CVSS score and LLM-typical, textbook-style formatting.

Hermes AIOS is an open-source agentic operating system built for autonomous AI with a focus on adaptability and user control. Combining the Hermes Agent with the ION UI, it adds long-term memory and reusable skills — so agents can actually pick up on how you work.

Most multi-agent systems fail the same way: agents drift apart across handoffs. By turn 3 they are working in different realities. By turn 5 they are repeating each other's mistakes and calling it parallelism.

Cyber Command plans to test and deploy the strongest available AI models regardless of politics or country of origin, chief AI officer Brig. Reid Novotny told Axios. While the White House negotiates Anthropic Mythos access, Cyber Command is building infrastructure that can swap models from any vendor — including open-source or boutique Chinese systems.

Malicious actors are now exploiting generative AI to carry out cyberattacks: scamming victims using AI-generated deepfakes, deploying malware developed with the help of AI coding tools, using chatbots for phishing, and hacking widely used open-source code repositories with AI agents. Anthropic's Frontier Red Team announced that the company's Claude Mythos Preview model has identified thousands of high- and critical-severity vulnerabilities, including so…

Transcribing audio to text on your PC is made accessible and secure with Vibe, an open source application that operates entirely offline. By using OpenAI’s Whisper model, Vibe supports transcription in multiple languages and handles challenging scenarios, such as audio with background noise or rapid speech.

The Hermes Agent, developed by Noose Research, is an open source AI system designed to enhance workflows and assist collaboration with large language models (LLMs). It incorporates features such as persistent memory, automated skill generation, and iterative learning to address complex tasks.

DeepSeek 4 introduces two open source language models designed to meet varying computational requirements, as detailed by Prompt Engineering. The Pro model, with 1.6 trillion parameters, is optimized for tasks demanding high precision and processing power, while the Flash model, featuring 284 billion parameters, is suited for environments with limited resources.

OpenClaw is an open source AI agent designed to act as a fully autonomous “AI employee,” handling tasks such as coding, research and device control. Alex Finn outlines the setup process, emphasizing the importance of using personal devices or dedicated machines instead of Virtual Private Servers (VPS).

Amazon announced the availability of G7e instances powered by NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs on Amazon SageMaker AI. You can provision nodes with 1, 2, 4, and 8 GPU instances, with each GPU providing 96 GB of GDDR7 memory. This enables cost-effective hosting of large foundation models like GPT-OSS-120B and Qwen3.5-35B-A3B on a single node.

A developer built an MCP server for the Novation Circuit Tracks, a hardware device for electronic music. The server gives an AI agent tools to compose and play music directly on the hardware. Users can describe what they want — 'a melodic ambient song with a dark atmosphere' — and the AI executes it.

The “Superpowers” plugin, created by Jesse Vincent, introduces a structured workflow that significantly enhances cloud code development. Nate Herk explores how this open source solution automates key phases like brainstorming, planning and debugging to improve productivity and code quality.